maskrcnn_benchmark.cpp

faster rcnn cpp implementation based on maskrcnn-benchmark

Codes

All code architecture from maskrcnn-benchmark

Installation

export INSTALL_DIR=$PWD
git clone --recursive https://github.com/lsrock1/maskrcnn_benchmark.cpp.git
cd $INSTALL_DIR

// if opencv is not installed
// opencv install
cd maskrcnn_benchmark.cpp/lib/opencv
mkdir build
cd build
cmake ..
make install

// if yaml-cpp is not installed
// yaml-cpp install
cd $INSTALL_DIR
cd maskrcnn_benchmark.cpp/lib/yaml-cpp
mkdir build
cd build
cmake ..
make install

// Download pytorch cpp (cxx11 ABI)
// place libtorch folder in maskrcnn_benchmark.cpp/lib directory
// if you are using without cuda and get dyld: Library not loaded: @rpath/libmklml.dylib error
// https://github.com/pytorch/pytorch/issues/14165

cd $INSTALL_DIR
cd maskrcnn_benchmark.cpp
mkdir build
cd build
cmake ..
make

//run inference r50-fpn
run.out ../configs/e2e_faster_rcnn_R_50_FPN_1x.yaml inference

Datasets

Download coco datasets
make directory under maskrcnn_benchmark.cpp

datasets
  - coco
    - train2017
    - val2017
  - annotations
    - instances_val2017.json
    - instances_train2017.json

Results

Device:  RTX2080ti 1 GPU, cuda 10, cudnn 7, Ubuntu 16.04

ResNet

VoVNet

ResNets are 1x and VoVNets are 2x

Warning

In Progress.

• It doesn't support training yet.(Testing!)

TODO

• concat dataset

• python jit -> cpp model code

• multi GPU training(code complete but bug exists in libtorch)

• cmake install

• clean up code

Requirements

• Yaml-cpp

• gtest

• libtorch >= 1.2

• rapidjson

• opencv

MODELS

Download into /models